CN108897042A - Organic matter content earthquake prediction method and device - Google Patents
Organic matter content earthquake prediction method and device Download PDFInfo
- Publication number
- CN108897042A CN108897042A CN201810986762.1A CN201810986762A CN108897042A CN 108897042 A CN108897042 A CN 108897042A CN 201810986762 A CN201810986762 A CN 201810986762A CN 108897042 A CN108897042 A CN 108897042A
- Authority
- CN
- China
- Prior art keywords
- network model
- organic matter
- content
- velocity
- log data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 239000005416 organic matter Substances 0.000 title claims abstract description 150
- 238000000034 method Methods 0.000 title claims abstract description 65
- 238000012549 training Methods 0.000 claims abstract description 89
- 238000013528 artificial neural network Methods 0.000 claims abstract description 44
- 239000000203 mixture Substances 0.000 claims abstract 8
- 210000005036 nerve Anatomy 0.000 claims description 61
- 230000006870 function Effects 0.000 claims description 23
- 210000002569 neuron Anatomy 0.000 claims description 21
- 238000009826 distribution Methods 0.000 claims description 20
- 238000004422 calculation algorithm Methods 0.000 claims description 18
- 238000012360 testing method Methods 0.000 claims description 15
- 238000012545 processing Methods 0.000 claims description 12
- 238000003860 storage Methods 0.000 claims description 9
- 238000004590 computer program Methods 0.000 claims description 6
- 238000009499 grossing Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 3
- 239000011435 rock Substances 0.000 description 13
- 230000000694 effects Effects 0.000 description 12
- 238000010586 diagram Methods 0.000 description 10
- 238000004458 analytical method Methods 0.000 description 6
- 238000013507 mapping Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 238000011160 research Methods 0.000 description 5
- 238000000465 moulding Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 238000003062 neural network model Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 241000208340 Araliaceae Species 0.000 description 2
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 2
- 235000003140 Panax quinquefolius Nutrition 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 239000002131 composite material Substances 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 235000008434 ginseng Nutrition 0.000 description 2
- 241000212384 Bifora Species 0.000 description 1
- 241001269238 Data Species 0.000 description 1
- 241001465754 Metazoa Species 0.000 description 1
- 238000000342 Monte Carlo simulation Methods 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000012790 confirmation Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000005284 excitation Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 239000003208 petroleum Substances 0.000 description 1
- 238000003825 pressing Methods 0.000 description 1
- 238000010206 sensitivity analysis Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000000946 synaptic effect Effects 0.000 description 1
- 238000010998 test method Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
- G01V1/28—Processing seismic data, e.g. for interpretation or for event detection
- G01V1/30—Analysis
- G01V1/306—Analysis for determining physical properties of the subsurface, e.g. impedance, porosity or attenuation profiles
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V2210/00—Details of seismic processing or analysis
- G01V2210/60—Analysis
- G01V2210/62—Physical property of subsurface
- G01V2210/624—Reservoir parameters
Landscapes
- Engineering & Computer Science (AREA)
- Remote Sensing (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Acoustics & Sound (AREA)
- Environmental & Geological Engineering (AREA)
- Geology (AREA)
- General Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Geophysics (AREA)
- Geophysics And Detection Of Objects (AREA)
Abstract
The invention provides an organic matter content earthquake prediction method and device, wherein the method comprises the following steps: acquiring seismic elasticity parameters from logging data to be predicted; inputting the earthquake elastic parameters into a mixed Gaussian probability density network model generated by pre-training, and predicting the organic matter content corresponding to the earthquake elastic parameters; the Gaussian mixture probability density network model comprises a nonlinear relation between earthquake elastic parameters generated by pre-training and organic matter content; the Gaussian mixture probability density network model comprises an artificial neural network model and a Gaussian mixture model; and the output vector of the artificial neural network model is the input vector of the Gaussian mixture model. According to the technical scheme, the organic matter content is predicted based on data driving, so that the accuracy and efficiency of organic matter content prediction are improved.
Description
Technical field
The present invention relates to Earthquake Reservoir field, in particular to a kind of content of organic matter earthquake prediction method and dress
It sets.
Background technique
With the development of seismic exploration technique, the investigation of petroleum resources has had occurred from structure type oil gas and has been hidden within complexity
The obvious transformation of lithologic deposit.For complex lithology oil-gas reservoir, reservoir physical parameter is in reservoir prediction and evaluation, oil gas
Very important foundation is had become in the determination of reserve estimate and development wells, it would therefore be desirable to consider it is more acurrate, more have
Effect, more economical method predict reservoir parameter.
Currently, the rock physics inversion method of predicting reservoir parameter mainly has based on petrophysical model from elastic parameter
Model driven method, which passes through various linear iteraction inversion algorithms or global sampling algorithm (Ru Mengte
Carlow sampling) it realizes, but this inversion method is limited to the petrophysical model constructed by us, for any geology ring
Rock physics relations under border are often semiempirical, need us using local log data and laboratory core measurement number
Petrophysical model according to being corrected, and after correcting also is only used for some geological conditions similar to model hypothesis condition
Under inverting, be difficult to solve the problems, such as the reservoir parameter forecast in any rock physics relations.Therefore, existing to be based on rock physics mould
There are limitation, the precision and low efficiency of prediction for the model-driven prediction technique of type.
In view of the above-mentioned problems, currently no effective solution has been proposed.
Summary of the invention
The embodiment of the invention provides a kind of content of organic matter earthquake prediction methods, to pass through having based on data-driven
Machine matter content prediction, improves the precision and efficiency of content of organic matter prediction, and this method includes：
Seismic elastic parameter is obtained from log data to be predicted；
The seismic elastic parameter is input in the mixed Gaussian probability density network model that training generates in advance, prediction
The corresponding content of organic matter of seismic elastic parameter；The mixed Gaussian probability density network model includes the ground that preparatory training generates
Shake the non-linear relation of elastic parameter and the content of organic matter；The mixed Gaussian probability density network model includes artificial neural network
Network model and gauss hybrid models；The output vector of the artificial nerve network model be the gauss hybrid models input to
Amount.
The embodiment of the invention also provides a kind of content of organic matter earthquake prediction apparatus, to by based on data-driven
Content of organic matter prediction, improves the precision and efficiency of content of organic matter prediction, which includes：
Acquiring unit, for obtaining seismic elastic parameter from log data to be predicted；
Predicting unit, for the seismic elastic parameter to be input to the mixed Gaussian probability density net that training generates in advance
In network model, the corresponding content of organic matter of prediction seismic elastic parameter；The mixed Gaussian probability density network model includes pre-
The first non-linear relation of the seismic elastic parameter that training generates and the content of organic matter；The mixed Gaussian probability density network model
Including artificial nerve network model and gauss hybrid models；The output vector of the artificial nerve network model is mixed for the Gauss
The input vector of molding type.
The embodiment of the invention also provides a kind of computer equipments, including memory, processor and storage are on a memory
And the computer program that can be run on a processor, the processor execute the content of organic matter earthquake prediction method.
The embodiment of the invention also provides a kind of computer readable storage medium, the computer-readable recording medium storage
There is the computer program for executing content of organic matter earthquake prediction method.
Compared with the scheme in the prior art based on the driving of petrophysical model to predict the content of organic matter, the present invention
The technical solution that embodiment provides, firstly, seismic elastic parameter is obtained from log data to be predicted, then, by the earthquake
Elastic parameter is input in the mixed Gaussian probability density network model that training generates in advance, and prediction seismic elastic parameter is corresponding
The content of organic matter；Wherein, which includes the preparatory seismic elastic parameter for training generation and has
The non-linear relation of machine matter content, and then the content of organic matter predicted according to the non-linear relation, independent of rock physics mould
Type, therefore have wider applicability and practicability under complex geological condition；Again because of the mixed Gaussian probability density network mould
Type includes artificial nerve network model and gauss hybrid models, and the output vector of artificial nerve network model is gauss hybrid models
Input vector, need not rely on Monte carlo algorithm solve inversion solution Bayes posterior probability distribution, overcome Meng Teka
Limitation of the Lip river method in computational efficiency can be distributed with the Bayes posterior probability of arbitrary accuracy solving model parameter, therefore, base
The distribution form of mixed Gauss model is solved in neural network algorithm, and as Bayes posterior probability, this method is bright
It is aobvious to have saved calculating cost, computational efficiency is improved, to improve the precision and efficiency of content of organic matter prediction.
Detailed description of the invention
The drawings described herein are used to provide a further understanding of the present invention, constitutes part of this application, not
Constitute limitation of the invention.In the accompanying drawings：
Fig. 1 is the structural schematic diagram of mixed Gaussian probability density network model in the embodiment of the present invention；
Fig. 2 is the signal of the artificial nerve network model in the embodiment of the present invention in mixed Gaussian probability density network model
Figure；
Fig. 3 is the schematic diagram of the gauss hybrid models in the embodiment of the present invention in mixed Gaussian probability density network model；
Fig. 4 is the flow diagram of content of organic matter earthquake prediction method in the embodiment of the present invention；
Fig. 5 is sensitivity analysis of the content of organic matter TOC to density, velocity of longitudinal wave and shear wave velocity in the embodiment of the present invention
Schematic diagram；
Fig. 6 is statistics with histogram analysis and the Gauss Distribution Fitting knot of the log data content of organic matter in the embodiment of the present invention
Fruit schematic diagram；
Fig. 7 is the mixed Gaussian probability density network model obtained in the embodiment of the present invention using training, to log data
Carry out the content of organic matter schematic diagram that inverting obtains；
Fig. 8 is the structural schematic diagram of content of organic matter earthquake prediction apparatus in the embodiment of the present invention.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer, right below with reference to embodiment and attached drawing
The present invention is described in further details.Here, exemplary embodiment and its explanation of the invention is used to explain the present invention, but simultaneously
It is not as a limitation of the invention.
Inventor's discovery：The rock physics inversion method of predicting reservoir parameter mainly has based on rock object from elastic parameter
Manage the model driven method of model.Model-driven inverting passes through various linear iteraction inversion algorithms or global sampling algorithm (such as
Monte Carlo) it realizes, but this inversion method is limited to the petrophysical model constructed by us, for arbitrarily
Rock physics relations under matter environment are often semiempirical, us is needed to survey using local log data and laboratory core
Amount data are corrected, and the petrophysical model after correction is also only used for some geology similar to model hypothesis condition
In the case of inverting, be difficult to solve the problems, such as the reservoir parameter forecast in any rock physics relations.Therefore, existing to be based on rock object
There are limitation, the precision and low efficiency of prediction for the model-driven prediction technique of reason model.
Therefore, because inventor has found above-mentioned technical problem, propose a kind of based on actual seismic data and well logging number
According to data-driven method, data-driven inverting can preferably solve the problems, such as this for model-driven inverting, it
Elastic parameter space is directly extracted from actual seismic data or log data to the mapping relations in reservoir parameter space, is not depended on
In petrophysical model, therefore there is wider applicability under complex geological condition.Inventor is described below by long-term big
Amount experiment finds that the process of model used in the content of organic matter earthquake prediction method based on data-driven is introduced.
Artificial neural network is a kind of a kind of reflecting based on data-driven for imitating animal nerve network behavior latent structure
Model is penetrated, it can realize any complex nonlinear mapping relations for being difficult to be described with mathematical model by the study to data.
Some scholars apply the common-shot-gather of Neural Network Inversion composite traces, have obtained seimic wave velocity model.Some scholars mention
The method for solving of the one-dimensional indirect problem in conjunction with neural network and probability theory is gone out.Some scholars propose the neural network of regularization
Inversion method can steadily predict porosity from 3D seismic data.Some scholars calculate using adaptive backpropagation
Method trains neural network, defines lithologic interface according to the density of log data, neutron porosity and gamma logging result.Some
Scholar proposes the neural network method more typically changed, and can find out the conditional probability density distribution of discrete random variable,
They predict the lithofacies that this method is applied to composite traces.Mixing probability density network (MDN) is that a kind of multidimensional ﹑ is special
Neural network, its initial concept are to be proposed by the famous computer scientist Bishop professor of Britain in 1994.As
A kind of new probability inversion method, mixing probability density network basic thought be by the mixed model of multiple gaussian kernel functions with
BP feedforward network combines, and can be distributed with the Bayes posterior probability of arbitrary accuracy solving model parameter, i.e., about output variable t
Uncertainty estimation can indicate that structural schematic diagram is as shown in Figure 1 by its conditional probability density p (t | x).It is close in mixing probability
Spend network in, BP network full name be the artificial neural network based on error backpropagation algorithm, structure include an input layer,
One output layer and at least one hidden layer, for neuron models as shown in Fig. 2, x is input vector in Fig. 2, w is network connection
Synaptic weight, b be neural network node biasing, f is excitation function, and y is the output valve of neuron.
Some scholars propose diagonal mixing probability density network, i.e. the off diagonal element of covariance matrix is zero, and
Diagonal entry is the variance of each model parameter, and variance is not necessarily equal, and diagonally mixing probability density network, (Gauss is mixed
Molding type) be applied to log data porosity and shale content solution, the results showed that diagonal mixing probability density network exists
Preferable effect can be obtained in the inverting of the two parameters.The structure of diagonal mixing probability density network is as shown in figure 3, Fig. 3
In, αiFor the mixed coefficint of i-th of kernel function,It indicates i-th of diagonal gaussian kernel function, expresses shape
Formula is：
Wherein, c indicates target variable t=(t1,...,tc) dimension, μikIn the mean value vector for indicating i-th of kernel function
K-th of element, σikIndicate k-th of diagonal element in the covariance matrix of i-th of core.Therefore, the mean value of i-th kernel function and side
Difference can be expressed as μi=(μi1,...,μic) and ∑i=diag (σi1,...σic).For the diagonal element for guaranteeing covariance matrix
Element value is effective, must be requested that variances sigmaikFor nonnegative number, therefore the output z of corresponding BP network need to be done exponential transform；Meanwhile for
Mean μik, can directly be indicated with the output of corresponding BP network；In addition, mixed coefficint must satisfySpecific table
Existing parametric form is as follows：
From the above, it is seen that predicting the content of organic matter of fine and close oil and gas reservoir using petrophysical model, there are phases
When big difficulty, comparatively, the diagonal mixed Gaussian probability density network based on data-driven is one effective anti-
Drill method.
Based on the above, the present invention is directed to apply diagonal mixed Gaussian probability density network, effectively from log data
The middle nonlinearity relationship obtained between elastic parameter and the content of organic matter, and then can be predicted from seismic elastic parameter
The content of organic matter of reservoir.And reach following purpose：
1, the relationship between the elastic parameter and the content of organic matter of rock is considerably complicated, and there is no very perfect at present
Theoretical formula it is portrayed, therefore scheme provided in an embodiment of the present invention has from elastic parameter and the content of organic matter
Log data in obtain the ability of relationship between the two；
2, the non-linear relation between elastic parameter and the content of organic matter stored according to neural network, the present invention are implemented
The scheme that example provides has the ability that the content of organic matter is predicted from the elastic parameter that seismic data inversion obtains.
That is, the present embodiments relate to a kind of content of organic matter earthquake prediction sides based on mixed Gaussian probability density network
Method, it is complex in geological environment, when petrophysical model is difficult to portray the relationship between elastic parameter and the content of organic matter, mention
The original creation technology based on diagonal mixed Gaussian probability density network out is that elasticity ginseng can be effectively obtained from log data
Nonlinearity relationship between the several and content of organic matter, and then the organic matter of reservoir can be predicted from seismic elastic parameter
Content.
The difficult point of the invention is to train the preferable mixed Gaussian probability density network of generalization ability, from Earthquake Resilient
The content of organic matter is accurately predicted in parameter, improves the precision and efficiency of content of organic matter prediction.The method is characterized in that setting
It counts reasonable neural network structure and the input data of neural network is effectively pre-processed.
The technical solution of the embodiment of the present invention mainly includes following components：
1, the Statistical Distribution Characteristics of the log data and the content of organic matter of analysing elastic parameter and the content of organic matter.This
The effect of research contents is：(1) elastic parameter sensitive to the content of organic matter is analyzed, determines the input parameter of neural network；
(2) corresponding relationship of the content of organic matter and elastic parameter is analyzed, does basis for the prediction result of the following explanations content of organic matter；
(3) Statistical Distribution Characteristics for analyzing the content of organic matter, primarily determine the number of mixed Gaussian kernel function and the knot of neural network
Structure.
2, the log data of elastic parameter and the content of organic matter is smoothed, while to mixed Gaussian probability density
The input data of network is normalized.The effect of this research contents is：(1) smoothing effect may insure that training obtains
Neural network can match the scale of the elastic parameter that seismic inversion obtains, this step requires smoothly to fit to log data
When carrying out smoothly to be trained by log data to log data under the premise of retaining the high-frequency information of log data as far as possible
Obtained neural network can reflect the relationship between seismic elastic parameter and the content of organic matter；(2) to the normalizing of input data
Change processing can normalize the predictive ability final to neural network and have a great impact, therefore with the convergence of accelerans network
It needs to carry out test of many times careful selection and suitably normalizes mode.
3, reasonable neural network structure is designed, and reasonable mixed Gaussian probability density net is obtained by log data training
Network.The effect of this research contents is：Both acquisitions are trained by the log data to elastic parameter and the content of organic matter
Between non-linear relation, and save it in mixed Gaussian probability density network, this is most crucial one in entire patent
Step, the neural network that this step requires can not only carry out inverting to log data, and also wanting can be from seismic elastic parameter
In be finally inversed by the content of organic matter, that is to say, that the obtained neural network of training should have preferable generalization ability.
4, predict have from the elastic parameter that seismic inversion obtains according to the mixed Gaussian probability density network that training obtains
Machine matter content.The effect of this research contents is：On the basis of seismic elastic parameter is normalized, as
The input of neural network carries out inverting by the mixed Gaussian probability density network obtained before, and obtaining fine and close oil and gas reservoir has
The prediction result of machine matter content.
It describes in detail below to the content of organic matter earthquake prediction scheme based on mixed Gaussian probability density network
It is as follows.
Fig. 4 is the flow diagram of content of organic matter earthquake prediction method in the embodiment of the present invention, as shown in figure 4, the party
Method includes the following steps：
Step 101：Seismic elastic parameter is obtained from log data to be predicted；
Step 102：Seismic elastic parameter is input in the mixed Gaussian probability density network model that training generates in advance,
Predict the corresponding content of organic matter of seismic elastic parameter；Mixed Gaussian probability density network model includes according to multiple log datas
The seismic elastic parameter of sample training generation in advance and the non-linear relation of the content of organic matter；Mixed Gaussian probability density network mould
Type includes artificial nerve network model and gauss hybrid models；The output vector of artificial nerve network model is gauss hybrid models
Input vector.
Compared with the scheme in the prior art based on the driving of petrophysical model to predict the content of organic matter, the present invention
The technical solution that embodiment provides, firstly, seismic elastic parameter is obtained from log data to be predicted, then, by the earthquake
Elastic parameter is input in the mixed Gaussian probability density network model that training generates in advance, and prediction seismic elastic parameter is corresponding
The content of organic matter；Wherein, which includes the preparatory seismic elastic parameter for training generation and has
The non-linear relation of machine matter content, and then the content of organic matter predicted according to the non-linear relation, independent of rock physics mould
Type, therefore have wider applicability and practicability under complex geological condition；Again because of the mixed Gaussian probability density network mould
Type includes artificial nerve network model and gauss hybrid models, and the output vector of artificial nerve network model is gauss hybrid models
Input vector, need not rely on Monte carlo algorithm solve inversion solution Bayes posterior probability distribution, overcome Meng Teka
Limitation of the Lip river method in computational efficiency can be distributed with the Bayes posterior probability of arbitrary accuracy solving model parameter, therefore, base
The distribution form of mixed Gauss model is solved in neural network algorithm, and as Bayes posterior probability, this method is bright
It is aobvious to have saved calculating cost, computational efficiency is improved, to improve the precision and efficiency of content of organic matter prediction.
Below to the present embodiments relate to the step of describe in detail it is as follows.
(1) the step of introducing training mixed Gaussian probability density network model in advance first.
In one embodiment, it can train in advance as follows and generate the mixed Gaussian probability density network mould
Type：
Obtain log data sample；The log data sample includes density, velocity of longitudinal wave and shear wave velocity and organic matter
Content log data；
The log data sample is divided into training set and test set；
The artificial nerve network model is trained using the training set, determines the first of artificial nerve network model
Structure is walked, and determines the preliminary structure of gauss hybrid models；
According to the test set to the preliminary structure of the artificial nerve network model and the preliminary knot of gauss hybrid models
Structure is adjusted, and determines the mixed Gaussian probability density network model that the preparatory training generates.
1, the process of above-mentioned acquisition log data sample is introduced first：
Firstly, to analyze log data sample, obtain to the content of organic matter after getting log data sample
Sensitive elastic parameter (such as：Density, velocity of longitudinal wave and shear wave velocity determine the input parameter of neural network), in addition, analysis
The corresponding relationship of the content of organic matter and elastic parameter does basis for the prediction result of the following explanations content of organic matter.
In one embodiment, above-mentioned content of organic matter earthquake prediction method can also include：As follows to instruction
Practice collection to be pre-processed：
The log data of density, velocity of longitudinal wave, shear wave velocity and the content of organic matter is carried out according to seismological observation scale flat
Sliding processing；
The log data of density, velocity of longitudinal wave and shear wave velocity after smoothing processing is carried out at linear normalization respectively
Reason, and using the log data of density, velocity of longitudinal wave and shear wave velocity after normalized as artificial nerve network model
Input vector；Output vector of the content of organic matter as artificial nerve network model after being smoothed.
When it is implemented, being pre-processed to log data sample, after getting log data sample to elastic parameter
It is smoothed with the log data of the content of organic matter, while the input data of mixed Gaussian probability density network is returned
One change processing.The effect of this research contents is：(1) neural network that smoothing effect may insure that training obtains can be matchingly
The scale for the elastic parameter that shake inverting obtains, this step require retaining well logging number as far as possible to the smoothly appropriate of log data
According to high-frequency information under the premise of, log data smoothly allow anti-by the obtained neural network of log data training
Reflect the relationship between seismic elastic parameter and the content of organic matter；It (2) can be with accelerans net to the normalized of input data
The convergence of network normalizes the predictive ability final to neural network and has a great impact, it is therefore desirable to it is prudent to carry out test of many times
The suitable normalization mode of selection.
2, it secondly introduces according to pretreated acquisition log data sample, training mixed Gaussian probability density network model
Process：
When it is implemented, the design and training process of mixed Gaussian probability density network may include：
(1) according to the analysis of the first step in technical solution as a result, by pre- place is made to the sensitive elastic parameter of the content of organic matter
Input neural network after reason, and output of the smoothed out content of organic matter as neural network.
(2) activation primitive for designing hidden layer neuron in BP network can be tanh type transmission function, output layer
The activation primitive of neuron can be linear transfer function.
(3) weight of neural network is initialized, determines the maximum frequency of training and Optimal Parameters of network.
(4) according to the analysis to content of organic matter statistical nature, the number of gaussian kernel function is primarily determined, empirically really
Determine the number of hidden layer neuron.
(5) input data forward direction transmits to obtain output data, calculates output data and the output of desired target (after i.e. smooth
The content of organic matter of log data) between error, if error be greater than target error, to error back propagation calculate hidden layer
The weight modification amount and threshold modifying amount of neuron.
(6) (5) step is repeated, until the error sum of squares of network reaches minimum, the training of network terminates.
(7) (4)-(5) step is repeated, further determines that suitable kernel function number and hidden with test by repeatedly training
Neuron number containing layer.
In one embodiment, the artificial nerve network model is trained using the training set, is determined artificial
The preliminary structure of neural network model, and determine the preliminary structure of gauss hybrid models, may include：
The weight of artificial nerve network model is initialized according to the following method：It is mixed to Gauss using K average algorithm
Molding type assigns initial value, and the initial value of gauss hybrid models is assigned to the output layer threshold value of artificial nerve network model, and zero is attached
Close random number assigns other layers of artificial nerve network model as initial weight；
According to the Statistical Distribution Characteristics of the content of organic matter, the number of the optimal kernel function of the gauss hybrid models is determined；
It is corresponding artificial when the sum of artificial nerve network model weight is less than 1/10th of training set sample size
The hidden layer neuron number of neural network model is the quantity of hidden layer neuron；
The determining and matched artificial mind of training set sample size can be with the optimum training number of network model.
Based on this, when specific implementation the technical essential of diagonal mixed Gaussian probability density network training may include：
(1) reasonable for the initialization of neural network weight.Since the calculated result of neural network is dependent on initial power
Value, and its enter the especially fast therefore excessive initial weight of nonlinear area by will lead to mistake as a result, and too small initial
Weight will need more times to learn, so that neural network adapts to non-linear process.Therefore, usually that zero is attached
Initialization of the close random number as weight, but in view of the unconditional of target variable is distributed, it can use here a kind of more excellent
Weight initialization mode.Assuming that target variable is in Gaussian Profile, mainly using K average algorithm by giving a gauss hybrid models
Assign initial value, and this model parameter be assigned to network output layer threshold value, the weighting parameter of other layers be still initialized as near zero with
Machine number.Experiments have shown that the method can greatly reduce the training time of network and avoid falling into minimum.
(2) number of kernel function depends on the form of simulated posterior probability Density Distribution, by increasing kernel function
Number, the form of expression of mixed Gaussian probability density and the matching degree of posterior probability can improve.However, excessive kernel function meeting
Lead to bigger calculation amount and longer training time.Therefore, the appropriate selection of kernel function number is generally required by continuous
Test is guaranteed within the reasonable training time to determine with this, best by the mixed weighting energy of Gaussian probability density kernel function
Portray the Posterior probability distribution of model parameter in ground.
(3) determination of hidden layer neuron number is usually empirical or test of many times as a result, test method is served as reasons
Less until mostly increasing performance of the hidden layer neuron quantity until neural network and no longer improving.For hidden layer neuron number
The selection of amount, by optimizing the generalization ability (referring to machine learning algorithm to the adaptability of new samples) of network, from training sample
This quantity speculates the quantity of hidden layer neuron, when the sum of neural network weight is less than 1/10th of training samples number
When, network has preferable generalization ability.
(4) generalization ability of neural network is the important symbol for measuring neural network performance quality.Not network training
Number is more, and result more can correctly reflect the mapping relations between input and output.This is because collected sample data
Noise has been usually contained, when frequency of training is excessive, network has been will lead to and learns more about containing noise data, store it more
" individual character " of each training sample, and the global feature that all samples are presented is masked, the generalization ability of network is influenced, thus
There is overfitting problem, i.e. network can accurately map the relationship between known sample, but cannot correctly show reflecting for unknown data
Penetrate result.Meanwhile training sample is inadequate can also cause poor fitting problem.Under normal conditions, optimal training time how is selected
Number is also to need the summing up experience in test of many times, and in practical applications, we should guarantee the training sample for having enough as far as possible
This, and certain mechanism (for example training terminates in advance) confirmation and the matched frequency of training of sample size are combined, improve the general of network
Change ability.
(2) the step of being predicted using the mixed Gaussian probability density network model that preparatory training generates secondly is introduced.
In one embodiment, seismic elastic parameter is obtained from log data to be predicted, may include：
Analyze seismic elastic parameter sensitive to the content of organic matter in log data to be predicted；The seismic elastic parameter
Including：Density, velocity of longitudinal wave and shear wave velocity；
Pretreatment is normalized to density, velocity of longitudinal wave and shear wave velocity；
The seismic elastic parameter is input in the mixed Gaussian probability density network model that training generates in advance, prediction
The corresponding content of organic matter of seismic elastic parameter may include：
Pretreated density, velocity of longitudinal wave and shear wave velocity will be normalized and be input to the mixing height that training generates in advance
In this probability density network model, the corresponding content of organic matter of prediction seismic elastic parameter.
When it is implemented, analyzing seismic elastic parameter sensitive to the content of organic matter in log data to be predicted, and right
Seismic elastic parameter：Pretreatment is normalized in density, velocity of longitudinal wave and shear wave velocity, and content of organic matter prediction can be improved
Efficiency and precision.
Below as one example in conjunction with attached drawing 5 to Fig. 7, how to be implemented with the specification present invention.
The specific steps of the embodiment of the present invention may include as follows：
1. pair density, velocity of longitudinal wave, shear wave velocity and the content of organic matter log data carried out according to seismological observation scale
It is appropriate smooth；
2. the log data of pair density, velocity of longitudinal wave and shear wave velocity carries out linear normalization processing respectively, and will processing
Input of the data afterwards as artificial nerve network model, previous step treated the content of organic matter is as artificial neural network mould
The output of type；
3. test determines reasonable artificial nerve network model structure, including the reasonable node in hidden layer of selection, Gauss
Frequency of training and the number of iterations during kernel function, artificial nerve network model right-value optimization, according to pretreated elasticity
Parameter and content of organic matter data are trained artificial nerve network model, and obtaining being capable of the rationally inverting well logging content of organic matter
Mixed Gaussian probability density network model；
4. inverting obtains density, velocity of longitudinal wave and shear wave velocity from seismic data, and returns according to log data
One factor changed respectively is normalized three seismic elastic parameters；
5. the seismic elastic parameter after normalized is input to the resulting mixed Gaussian probability density network mould of training
Type, to be finally inversed by the content of organic matter in reservoir.
Wherein, for the effect of fine and close oil and gas reservoir content of organic matter prediction as shown in following examples, fine and close oil and gas reservoir
Content of organic matter prediction：
For fine and close oil and gas reservoir, first according to the well logging data analysis content of organic matter to the sensibility of elastic parameter,
It has been determined that density, velocity of longitudinal wave and shear wave velocity can be used as the actual parameter of the inverting content of organic matter, from analysis result (see figure
Density is density in 5, Fig. 5, and Vs is shear wave velocity, and Vp is velocity of longitudinal wave) in it can be seen that density and p-and s-wave velocity can
Preferably distinguish the different contents of organic matter.Next analyzes the Statistical Distribution Characteristics of the content of organic matter, as shown in fig. 6, horizontal
Axis TOC is the content of organic matter, and the longitudinal axis is statistics frequency, and curve is Gauss Distribution Fitting as a result, can from analysis result in figure
To find out that the content of organic matter of this area and is Unimodal Distribution close to Gaussian Profile, thus with a gaussian kernel function come pair
It is portrayed.
According to the training result of artificial nerve network model, we carry out inverting, knot to the content of organic matter of log data
For fruit as shown in fig. 7, a figure left side is probability of the content of organic matter when each depth takes different value, colour code is probability value, takes probability
Maximum value is inversion result, the inversion result of right smooth value (solid line) and the content of organic matter for the well logging content of organic matter of figure
The comparing result of (dotted line).
It can see by the inversion result to log data and seismic data, mixed Gaussian probability density network can have
Effect ground predicts the content of organic matter of fine and close oil and gas reservoir from density, velocity of longitudinal wave and shear wave velocity.
Based on the same inventive concept, a kind of content of organic matter earthquake prediction apparatus is additionally provided in the embodiment of the present invention, such as
The following examples.The principle solved the problems, such as due to content of organic matter earthquake prediction apparatus and above-mentioned content of organic matter earthquake prediction
Method is similar, therefore the implementation of content of organic matter earthquake prediction apparatus can be with reference to above-mentioned content of organic matter earthquake prediction method
Implement, overlaps will not be repeated.Used below, the software of predetermined function may be implemented in term " module " or " module "
And/or the combination of hardware.Although device described in following embodiment is preferably realized with software, hardware or soft
The realization of the combination of part and hardware is also that may and be contemplated.
Fig. 8 is the structural schematic diagram of content of organic matter earthquake prediction apparatus in the embodiment of the present invention, as shown in figure 8, the dress
Set including：
Acquiring unit 02, for obtaining seismic elastic parameter from log data to be predicted；
Predicting unit 04, for the seismic elastic parameter to be input to the mixed Gaussian probability density that training generates in advance
In network model, the corresponding content of organic matter of prediction seismic elastic parameter；The mixed Gaussian probability density network model includes
The non-linear relation of the seismic elastic parameter that training generates and the content of organic matter in advance；The mixed Gaussian probability density network mould
Type includes artificial nerve network model and gauss hybrid models；The output vector of the artificial nerve network model is the Gauss
The input vector of mixed model.
In one embodiment, which can also include：Training unit, for instructing in advance
Practice and generates the mixed Gaussian probability density network model；The training unit includes：
Module is obtained, for obtaining log data sample；The log data sample includes density, velocity of longitudinal wave and shear wave
Speed and content of organic matter log data；
Division module, for the log data sample to be divided into training set and test set；
It primarily determines module, for being trained using the training set to the artificial nerve network model, determines people
The preliminary structure of artificial neural networks model, and determine the preliminary structure of gauss hybrid models；
Final determining module, for the preliminary structure and Gauss according to the test set to the artificial nerve network model
The preliminary structure of mixed model is adjusted, and determines the mixed Gaussian probability density network model that the preparatory training generates.
In one embodiment, described to primarily determine that module specifically can be used for：
The weight of artificial nerve network model is initialized according to the following method：It is mixed to Gauss using K average algorithm
Molding type assigns initial value, and the initial value of gauss hybrid models is assigned to the output layer threshold value of artificial nerve network model, and zero is attached
Close random number assigns other layers of artificial nerve network model as initial weight；
According to the Statistical Distribution Characteristics of the content of organic matter, the number of the optimal kernel function of the gauss hybrid models is determined；
It is corresponding artificial when the sum of artificial nerve network model weight is less than 1/10th of training set sample size
The hidden layer neuron number of neural network model is the quantity of hidden layer neuron；
The determining optimal frequency of training with the matched artificial nerve network model of training set sample size.
In one embodiment, above-mentioned content of organic matter earthquake prediction apparatus can also include：Pretreatment unit, for pressing
Training set is pre-processed according to following method：
The log data of density, velocity of longitudinal wave, shear wave velocity and the content of organic matter is carried out according to seismological observation scale flat
Sliding processing；
The log data of density, velocity of longitudinal wave and shear wave velocity after smoothing processing is carried out at linear normalization respectively
Reason, and using the log data of density, velocity of longitudinal wave and shear wave velocity after normalized as artificial nerve network model
Input vector；Output vector of the content of organic matter as artificial nerve network model after being smoothed.
In one embodiment, acquiring unit specifically can be used for：
Analyze seismic elastic parameter sensitive to the content of organic matter in log data to be predicted；The seismic elastic parameter
May include：Density, velocity of longitudinal wave and shear wave velocity；
Pretreatment is normalized to density, velocity of longitudinal wave and shear wave velocity；
The predicting unit is specifically used for：Pretreated density, velocity of longitudinal wave and shear wave velocity input will be normalized
In the mixed Gaussian probability density network model generated to preparatory training, the corresponding content of organic matter of prediction seismic elastic parameter.
The embodiment of the invention also provides a kind of computer equipments, including memory, processor and storage are on a memory
And the computer program that can be run on a processor, the processor execute the content of organic matter earthquake prediction method.
The embodiment of the invention also provides a kind of computer readable storage medium, the computer-readable recording medium storage
There is the computer program for executing content of organic matter earthquake prediction method.
The advantageous effects of technical solution that the present invention implements to provide are：
1, the present invention implement provide technical solution can complete elastic parameter to the content of organic matter Nonlinear Mapping, from
And the content of organic matter in reservoir rock can be predicted.
2, the technical solution that the present invention implements to provide has higher resolution ratio independent of petrophysical model.Elasticity ginseng
It is a kind of extremely complex non-linear relation between the several and content of organic matter, it is carried out using petrophysical model to portray presence
Significant limitation, firstly, petrophysical model all has various assumed conditions, these are assumed in complicated geology feelings
It may and be not suitable under condition；Secondly, the petrophysical model portrayed at present to the content of organic matter is most of too simple, nothing
Method meets the required precision of reservoir content of organic matter prediction.And the present invention implements the general using mixing of the technical solution provided proposition
It is a kind of prediction technique based entirely on data-driven that rate density network, which carries out prediction to the content of organic matter, can utilize well logging number
According to Nonlinear Mapping complicated between completion elastic parameter and the content of organic matter, thus this method has more than petrophysical model
High resolution ratio, and for petrophysical model, under complex geological condition, the practicability of this method is obviously wanted
It is higher.
After 3. the technical solution that the present invention implements to provide needs not rely on the Bayes that Monte carlo algorithm solves inversion solution
Probability distribution is tested, limitation of the Monte Carlo method in computational efficiency is overcome.Due to the nonuniqueness of indirect problem, we are not only
Need to obtain the solution of indirect problem, it is also necessary to evaluate the uncertainty of solution, however pattra leaves is solved based on Monte carlo algorithm
The Posterior probability distribution calculation amount of this indirect problem is huge, and benefit is lower in the calculating of practical problem, therefore is based on neural network
Algorithm solves the distribution form of mixed Gauss model, and as Bayes posterior probability, this method has obviously saved meter
It is counted as this, improves computational efficiency.
To sum up, the present invention implements the technical solution provided by the content of organic matter prediction based on data-driven, improves
The precision and efficiency of content of organic matter prediction.
Obviously, those skilled in the art should be understood that each module of the above-mentioned embodiment of the present invention or each step can be with
It is realized with general computing device, they can be concentrated on a single computing device, or be distributed in multiple computing devices
On composed network, optionally, they can be realized with the program code that computing device can perform, it is thus possible to by it
Store and be performed by computing device in the storage device, and in some cases, can be held with the sequence for being different from herein
The shown or described step of row, perhaps they are fabricated to each integrated circuit modules or will be multiple in them
Module or step are fabricated to single integrated circuit module to realize.In this way, the embodiment of the present invention be not limited to it is any specific hard
Part and software combine.
The foregoing is only a preferred embodiment of the present invention, is not intended to restrict the invention, for the skill of this field
For art personnel, the embodiment of the present invention can have various modifications and variations.All within the spirits and principles of the present invention, made
Any modification, equivalent substitution, improvement and etc. should all be included in the protection scope of the present invention.
Claims (12)
1. a kind of content of organic matter earthquake prediction method, which is characterized in that including：
Seismic elastic parameter is obtained from log data to be predicted；
The seismic elastic parameter is input in the mixed Gaussian probability density network model that training generates in advance, predicts earthquake
The corresponding content of organic matter of elastic parameter；The mixed Gaussian probability density network model includes the earthquake bullet that preparatory training generates
The non-linear relation of property parameter and the content of organic matter；The mixed Gaussian probability density network model includes artificial neural network mould
Type and gauss hybrid models；The output vector of the artificial nerve network model is the input vector of the gauss hybrid models.
2. content of organic matter earthquake prediction method as described in claim 1, which is characterized in that training in advance as follows
Generate the mixed Gaussian probability density network model：
Obtain log data sample；The log data sample includes density, velocity of longitudinal wave and shear wave velocity and the content of organic matter
Log data；
The log data sample is divided into training set and test set；
The artificial nerve network model is trained using the training set, determines the preliminary knot of artificial nerve network model
Structure, and determine the preliminary structure of gauss hybrid models；
According to the test set to the preliminary structure of the preliminary structure of the artificial nerve network model and gauss hybrid models into
Row adjustment determines the mixed Gaussian probability density network model that the preparatory training generates.
3. content of organic matter earthquake prediction method as claimed in claim 2, which is characterized in that using the training set to described
Artificial nerve network model is trained, and determines the preliminary structure of artificial nerve network model, and determine gauss hybrid models
Preliminary structure, including：
The weight of artificial nerve network model is initialized according to the following method：Gaussian Mixture mould is given using K average algorithm
Type assigns initial value, and the initial value of gauss hybrid models is assigned to the output layer threshold value of artificial nerve network model, near zero
Random number assigns other layers of artificial nerve network model as initial weight；
According to the Statistical Distribution Characteristics of the content of organic matter, the number of the optimal kernel function of the gauss hybrid models is determined；
When the sum of artificial nerve network model weight is less than 1/10th of training set sample size, corresponding artificial neuron
The hidden layer neuron number of network model is the quantity of hidden layer neuron；
The determining optimum training number with the matched artificial nerve network model of training set sample size.
4. content of organic matter earthquake prediction method as claimed in claim 2, which is characterized in that further include：As follows
Training set is pre-processed：
The log data of density, velocity of longitudinal wave, shear wave velocity and the content of organic matter is smoothly located according to seismological observation scale
Reason；
Linear normalization processing is carried out respectively to the log data of density, velocity of longitudinal wave and shear wave velocity after smoothing processing, and
Using the log data of density, velocity of longitudinal wave and shear wave velocity after normalized as the input of artificial nerve network model to
Amount；Output vector of the content of organic matter as artificial nerve network model after being smoothed.
5. content of organic matter earthquake prediction method as claimed in claim 2, which is characterized in that from log data to be predicted
Seismic elastic parameter is obtained, including：
Analyze seismic elastic parameter sensitive to the content of organic matter in log data to be predicted；The seismic elastic parameter packet
It includes：Density, velocity of longitudinal wave and shear wave velocity；
Pretreatment is normalized to density, velocity of longitudinal wave and shear wave velocity；
The seismic elastic parameter is input in the mixed Gaussian probability density network model that training generates in advance, predicts earthquake
The corresponding content of organic matter of elastic parameter, including：
Will be normalized pretreated density, velocity of longitudinal wave and shear wave velocity be input in advance training generate mixed Gaussian it is general
In rate density network model, the corresponding content of organic matter of prediction seismic elastic parameter.
6. a kind of content of organic matter earthquake prediction apparatus, which is characterized in that including：
Acquiring unit, for obtaining seismic elastic parameter from log data to be predicted；
Predicting unit, for the seismic elastic parameter to be input to the mixed Gaussian probability density network mould that training generates in advance
In type, the corresponding content of organic matter of prediction seismic elastic parameter；The mixed Gaussian probability density network model includes instructing in advance
Practice the non-linear relation of the seismic elastic parameter and the content of organic matter that generate；The mixed Gaussian probability density network model includes
Artificial nerve network model and gauss hybrid models；The output vector of the artificial nerve network model is the Gaussian Mixture mould
The input vector of type.
7. content of organic matter earthquake prediction apparatus as claimed in claim 6, which is characterized in that further include：Training unit is used for
Training generates the mixed Gaussian probability density network model in advance；The training unit includes：
Module is obtained, for obtaining log data sample；The log data sample includes density, velocity of longitudinal wave and shear wave velocity
With content of organic matter log data；
Division module, for the log data sample to be divided into training set and test set；
It primarily determines module, for being trained using the training set to the artificial nerve network model, determines artificial mind
Preliminary structure through network model, and determine the preliminary structure of gauss hybrid models；
Final determining module, for the preliminary structure and Gaussian Mixture according to the test set to the artificial nerve network model
The preliminary structure of model is adjusted, and determines the mixed Gaussian probability density network model that the preparatory training generates.
8. content of organic matter earthquake prediction apparatus as claimed in claim 7, which is characterized in that described to primarily determine that module is specific
For：
The weight of artificial nerve network model is initialized according to the following method：Gaussian Mixture mould is given using K average algorithm
Type assigns initial value, and the initial value of gauss hybrid models is assigned to the output layer threshold value of artificial nerve network model, near zero
Random number assigns other layers of artificial nerve network model as initial weight；
According to the Statistical Distribution Characteristics of the content of organic matter, the number of the optimal kernel function of the gauss hybrid models is determined；
When the sum of artificial nerve network model weight is less than 1/10th of training set sample size, corresponding artificial neuron
The hidden layer neuron number of network model is the quantity of hidden layer neuron；
The determining optimal frequency of training with the matched artificial nerve network model of training set sample size.
9. content of organic matter earthquake prediction apparatus as claimed in claim 7, which is characterized in that further include：Pretreatment unit is used
In being pre-processed as follows to training set：
The log data of density, velocity of longitudinal wave, shear wave velocity and the content of organic matter is smoothly located according to seismological observation scale
Reason；
Linear normalization processing is carried out respectively to the log data of density, velocity of longitudinal wave and shear wave velocity after smoothing processing, and
Using the log data of density, velocity of longitudinal wave and shear wave velocity after normalized as the input of artificial nerve network model to
Amount；Output vector of the content of organic matter as artificial nerve network model after being smoothed.
10. content of organic matter earthquake prediction apparatus as claimed in claim 7, which is characterized in that the acquiring unit is specifically used
In：
Analyze seismic elastic parameter sensitive to the content of organic matter in log data to be predicted；The seismic elastic parameter packet
It includes：Density, velocity of longitudinal wave and shear wave velocity；
Pretreatment is normalized to density, velocity of longitudinal wave and shear wave velocity；
The predicting unit is specifically used for：To be normalized pretreated density, velocity of longitudinal wave and shear wave velocity be input to it is pre-
In the mixed Gaussian probability density network model that first training generates, the corresponding content of organic matter of prediction seismic elastic parameter.
11. a kind of computer equipment including memory, processor and stores the meter that can be run on a memory and on a processor
Calculation machine program, which is characterized in that the processor realizes any side of claim 1 to 5 when executing the computer program
Method.
12. a kind of computer readable storage medium, which is characterized in that the computer-readable recording medium storage has perform claim
It is required that the computer program of 1 to 5 any the method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810986762.1A CN108897042B (en) | 2018-08-28 | 2018-08-28 | Organic matter content earthquake prediction method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810986762.1A CN108897042B (en) | 2018-08-28 | 2018-08-28 | Organic matter content earthquake prediction method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108897042A true CN108897042A (en) | 2018-11-27 |
| CN108897042B CN108897042B (en) | 2020-03-10 |
Family
ID=64358522
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810986762.1A Active CN108897042B (en) | 2018-08-28 | 2018-08-28 | Organic matter content earthquake prediction method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108897042B (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110118994A (en) * | 2019-05-24 | 2019-08-13 | 同济大学 | A kind of nonmarine source rock quantitative forecasting technique based on seismic inversion and machine learning |
| CN110161565A (en) * | 2019-03-20 | 2019-08-23 | 中国海洋石油集团有限公司 | A kind of Reconstruction of seismic data method |
| CN110954950A (en) * | 2019-10-31 | 2020-04-03 | 南方科技大学 | Underground transverse wave velocity inversion method, device, computing equipment and storage medium |
| CN111983515A (en) * | 2020-08-28 | 2020-11-24 | 广东电网有限责任公司广州供电局 | Voltage sag frequency random evaluation method and device |
| CN112444855A (en) * | 2019-08-30 | 2021-03-05 | 中国石油化工股份有限公司 | Seismic facies analysis method based on Gaussian mixture model and storage medium |
| CN112596102A (en) * | 2020-10-27 | 2021-04-02 | 中国石油天然气股份有限公司 | Geophysical prediction method, device and medium for oil reservoir seepage characteristic parameters |
| CN114114414A (en) * | 2021-11-18 | 2022-03-01 | 电子科技大学长三角研究院(湖州) | Artificial intelligence prediction method for 'dessert' information of shale reservoir |
| CN114529110A (en) * | 2020-11-03 | 2022-05-24 | 中国石油天然气集团有限公司 | Lithofacies inversion method and system based on deep neural network model |
| CN114910960A (en) * | 2021-02-09 | 2022-08-16 | 中国石油天然气股份有限公司 | Reservoir parameter prediction method and device |
| US12019202B2 (en) | 2021-03-25 | 2024-06-25 | Saudi Arabian Oil Company | Fast variogram modeling driven by artificial intelligence |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103760600A (en) * | 2014-01-07 | 2014-04-30 | 中国石油天然气股份有限公司 | Gas saturation inversion method |
| CN104749624A (en) * | 2015-03-03 | 2015-07-01 | 中国石油大学(北京) | Method for synchronously realizing seismic lithofacies identification and quantitative assessment of uncertainty of seismic lithofacies identification |
| CN106324674A (en) * | 2016-08-23 | 2017-01-11 | 中国石油大学(华东) | Shale gas TOC pre-stack seismic inversion prediction method |
-
2018
- 2018-08-28 CN CN201810986762.1A patent/CN108897042B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103760600A (en) * | 2014-01-07 | 2014-04-30 | 中国石油天然气股份有限公司 | Gas saturation inversion method |
| CN104749624A (en) * | 2015-03-03 | 2015-07-01 | 中国石油大学(北京) | Method for synchronously realizing seismic lithofacies identification and quantitative assessment of uncertainty of seismic lithofacies identification |
| CN106324674A (en) * | 2016-08-23 | 2017-01-11 | 中国石油大学(华东) | Shale gas TOC pre-stack seismic inversion prediction method |
Non-Patent Citations (2)
| Title |
|---|
| 晏信飞，等: "致密砂岩储层贝叶斯岩性判别与孔隙流体检测", 《石油地球物理勘探》 * |
| 殷习容，等: "基于混合概率密度网络的岩石物理反演", 《中国地球物理2013》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110161565A (en) * | 2019-03-20 | 2019-08-23 | 中国海洋石油集团有限公司 | A kind of Reconstruction of seismic data method |
| CN110161565B (en) * | 2019-03-20 | 2021-05-11 | 中国海洋石油集团有限公司 | Seismic data reconstruction method |
| CN110118994A (en) * | 2019-05-24 | 2019-08-13 | 同济大学 | A kind of nonmarine source rock quantitative forecasting technique based on seismic inversion and machine learning |
| CN112444855A (en) * | 2019-08-30 | 2021-03-05 | 中国石油化工股份有限公司 | Seismic facies analysis method based on Gaussian mixture model and storage medium |
| CN112444855B (en) * | 2019-08-30 | 2024-04-09 | 中国石油化工股份有限公司 | Seismic phase analysis method based on Gaussian mixture model and storage medium |
| CN110954950B (en) * | 2019-10-31 | 2022-10-21 | 南方科技大学 | Underground transverse wave velocity inversion method, device, computing equipment and storage medium |
| CN110954950A (en) * | 2019-10-31 | 2020-04-03 | 南方科技大学 | Underground transverse wave velocity inversion method, device, computing equipment and storage medium |
| CN111983515A (en) * | 2020-08-28 | 2020-11-24 | 广东电网有限责任公司广州供电局 | Voltage sag frequency random evaluation method and device |
| CN112596102B (en) * | 2020-10-27 | 2023-08-22 | 中国石油天然气股份有限公司 | Geophysical prediction method, device and medium for oil reservoir seepage characteristic parameters |
| CN112596102A (en) * | 2020-10-27 | 2021-04-02 | 中国石油天然气股份有限公司 | Geophysical prediction method, device and medium for oil reservoir seepage characteristic parameters |
| CN114529110A (en) * | 2020-11-03 | 2022-05-24 | 中国石油天然气集团有限公司 | Lithofacies inversion method and system based on deep neural network model |
| CN114910960A (en) * | 2021-02-09 | 2022-08-16 | 中国石油天然气股份有限公司 | Reservoir parameter prediction method and device |
| US12019202B2 (en) | 2021-03-25 | 2024-06-25 | Saudi Arabian Oil Company | Fast variogram modeling driven by artificial intelligence |
| CN114114414A (en) * | 2021-11-18 | 2022-03-01 | 电子科技大学长三角研究院(湖州) | Artificial intelligence prediction method for 'dessert' information of shale reservoir |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108897042B (en) | 2020-03-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108897042A (en) | Organic matter content earthquake prediction method and device | |
| Pham et al. | Missing well log prediction using convolutional long short-term memory network | |
| Mousavi et al. | Deep-learning seismology | |
| Liu et al. | Deep-learning seismic full-waveform inversion for realistic structural models | |
| US11397272B2 (en) | Data augmentation for seismic interpretation systems and methods | |
| US11320551B2 (en) | Training machine learning systems for seismic interpretation | |
| CN111551992B (en) | Rock reservoir structure characterization method and device, computer-readable storage medium and electronic equipment | |
| Grana et al. | Probabilistic inversion of seismic data for reservoir petrophysical characterization: Review and examples | |
| Guo et al. | Joint inversion of audio-magnetotelluric and seismic travel time data with deep learning constraint | |
| Wang et al. | Data-driven S-wave velocity prediction method via a deep-learning-based deep convolutional gated recurrent unit fusion network | |
| Wang et al. | Missing well logs prediction using deep learning integrated neural network with the self-attention mechanism | |
| Jeong et al. | A fast approximation for seismic inverse modeling: Adaptive spatial resampling | |
| CN116047583A (en) | Adaptive wave impedance inversion method and system based on depth convolution neural network | |
| Fang et al. | Deep generator priors for Bayesian seismic inversion | |
| Alzahrani et al. | Seismic velocity model building using neural networks: Training data design and learning generalization | |
| Zhang et al. | Extracting Q anomalies from marine reflection seismic data using deep learning | |
| Song et al. | Semi-supervised learning seismic inversion based on spatio-temporal sequence residual modeling neural network | |
| Hansen et al. | Use of machine learning to estimate statistics of the posterior distribution in probabilistic inverse problems—An application to airborne EM data | |
| You et al. | Explainable convolutional neural networks driven knowledge mining for seismic facies classification | |
| Naprstek et al. | Convolutional neural networks applied to the interpretation of lineaments in aeromagnetic data | |
| Fu et al. | Deep Conditional Generative Adversarial Network Combined With Data‐Space Inversion for Estimation of High‐Dimensional Uncertain Geological Parameters | |
| Liu et al. | Recurrent neural network for seismic reservoir characterization | |
| Shahriari et al. | A deep learning approach to design a borehole instrument for geosteering | |
| Qu et al. | Using a synthetic data trained convolutional neural network for predicting subresolution thin layers from seismic data | |
| Prezioso et al. | A data-driven artificial neural network model for the prediction of ground motion from induced seismicity: The case of The Geysers geothermal field |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |